[R-sig-Geo] sample.Spatial question

[Roger Bivand]

On Thu, 12 Nov 2009, Edzer Pebesma wrote:

> This is not a full answer. To start of with, I'd suggest to use spsample
> rather than sample.Spatial (maybe we should not make the latter
> available to usage).
>
> rand.points <- spsample(SPDF,n=23,type="hexagonal")
> plot(rand.points, add=T,col='red')
>
> The pattern you sketch below (nice ascii art!) might not fit in a square
> area.
>
> I'm afraid we overlooked this, and need to modify the documentation to
> state that only for a limited number of sampling methods for square
> regions the sample size is guaranteed the requested number. And exclude
> the hexagonal...

Yes, the outcome reported is based on the cellsize= argument not being 
given, so it defaults to:

area = prod(apply(bb, 1, diff))/n
dx = sqrt(area)/(sqrt(3)/2)

If you use:

rand.points <- spsample(SPDF, cellsize=80, offset=c(0.3, 0.6),
   type="hexagonal")

then

dim(coordinates(rand.points))

is

[1] 23  2

but it needs some playing around with both cellsize= and offset=. Using 
cellsize= rather than n= may be sensible. For sample.Polygon which is the 
method being used here, offset= defaults to runif(2), so should be set to 
a known value to anchor the grid.

Roger

> --
> Edzer
>
>
> Chen, Shaofei wrote:
>> Dear list members,
>>
>> Thanks to Dr. Bivand's quick reply on my last question about creating
>> a specific polygon. My next question follows that one.
>> After creating a polygon, I want to sample inside the polygon, I used
>> sample.Spatial in sp library. I want to sample 23 points in a
>> hexagonal structure.
>>
>> library(sp)
>> crds <- cbind(x=c(0, 0, 400, 400, 0), y=c(0, 400, 400, 0, 0))
>> Pl <- Polygon(crds)
>> ID <- "400x400"
>> Pls <- Polygons(list(Pl), ID=ID)
>> SPls <- SpatialPolygons(list(Pls))
>> df <- data.frame(value=1, row.names=ID)
>> SPDF <- SpatialPolygonsDataFrame(SPls, df)
>> plot(SPDF)
>> rand.points <- sample.Spatial(SPDF,n=23,type="hexagonal")
>> plot(rand.points, add=T)
>>
>>
>> According to my imagination, the point pattern should look like this:
>>
>> +   +  +  +  +
>>   +  +  +  +
>> +   +  +  +  +
>>   +  +  +  +
>> +   +  +  +  +
>>
>> However, everytime, the results are not 23 points, they usually come
>> up with 16 points.
>> I changed to regular sampling structure with a specified offset
>> c(0.5,0.5). It can give consistent results for 4x4=16 or 5x5=25...
>> points. But for other number of points, for example, 20 points, it
>> cannot give a 4x5 point pattern. Also offset option is not for other
>> sampling type.
>>
>> I just wonder why this happens. I look at the source code,  I think it
>> may because that the it defines the first point randomly (of course,
>> it uses "area = prod(apply(bb, 1, diff))/n" to define the stratum),
>> and then construct the following samples.  In this way, some points
>> may be out of the boundary.
>>
>> Is it possible that I can get exactly number of points that I want for
>> sampling?  And also I want the offset for hexagon is available, so
>> that I can make sure that the sampling points locating in the centroid
>> of the hexagon. I checked the archives of the list, Dr. Bivand has
>> provided codes that not utilize sample.Spatial  for creating hexagons
>> and centroids (I have done it with 23 points in 400 by 400 polygon),
>> but it may not be easy in unregular polygon with defined number of
>> points.
>>
>> Thank you in advance!
>>
>> Best,
>> Shaofei
>>
>> _______________________________________________
>> R-sig-Geo mailing list
>> R-sig-Geo at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no